GB/T 38643-2020 in English

Standard Overview and Technical Background

GB/T38643-2020, as a supporting test standard for GB/T37721-2019, establishes a functional test framework for big data analysis systems. The standard was jointly drafted by 15 units including Inspur Electronics and China Electronics Technology Standardization Institute, reflecting my country's technological accumulation in the field of big data analysis.

Core module test requirements

Details of key test items

1. Machine learning function test

The standard requires that the test system must support 6 types of core algorithms:

• Regression and classification algorithms: The ROC curves of models such as logistic regression and decision tree need to be verified
• Neural network algorithm: The recognition accuracy of the MNIST data set should be ≥ 95%
• Feature engineering component: The explained variance of the data after PCA dimensionality reduction needs to be tested to be ≥ 80%

2. Stream data analysis test

The standard defines three types of test scenarios:

1. Time window processing: need to verify the data integrity of the 1s~1h adjustable window
2. Multi-stream association: test the timeliness of JOIN operations of ≥2 data streams
3. State management: verify the fault recovery time under the checkpoint mechanism

Implementation suggestions

Test environment construction

It is recommended to use containerized deployment of the test environment. The resource configuration should meet the following requirements:

• Computing nodes: ≥8-core CPU/32GB memory
• Storage system: HDFS+Alluxio hybrid architecture
• Network bandwidth: ≥10Gbps

Automated testing framework

Recommended test tool combination: